Home / Articles / Ways to read one article: the finished piece stays at the centre, and the reader picks the way in

Ways to read one article: the finished piece stays at the centre, and the reader picks the way in

By Dinis Cruz · 2026-10-12 · article v1.0.1, 2 versions · site v0.7.70 · newsroomreadingviewsfive-readersinfographicsslidesgraphslocal-modelspersonasmobileemailbriefarticle

Abstract: One article, five readers worked: a Librarian, a Cartographer, a Historian, an Explainer and a Storyteller read five articles after they were written and produced the catalogue, the maps, the arc, the two-minute version and the deck. What did not work was the delivery. All of it landed as one more fold under the abstract, everything mixed, a wall of material next to a wall of text, and a reader with two minutes was no better off than before. This article is the next step, now that the pane on the right of every newsroom page is about the piece in front of the reader. The reader picks the way in, one tap, and can set a default: headlines, the infographic, two minutes, the slides, what is new against what they have read, the facts, the relationships, the vocabulary, or, through a model in their own browser, a view at their level and through their persona. Most of those are computed from views that already exist, deterministic and explainable; a local model is for the rest, and none of it round-trips to the publisher. And one thing is kept whole on purpose: the finished article. It has a beginning, a pacing, a surprise and a voice, and it is the one artefact that has to make sense from end to end, which a graph cannot. The modes are ways in, not replacements, and the brief says how to build them for a phone, a tablet, a desktop and an email.

In One article, five readers I sent five agents through five articles after they were written, each with a role, and they produced what I hoped for: a catalogue of every fact and claim with its sentence, an ontology and maps, an account of what each article added, a two-minute version and a deck. It worked. The views are on those articles today, on this site and on the newsroom.

What did not work was how they arrived. Open one of those articles and under the abstract there is a fold that says "five readers: two minutes, the arc, nine slides, a map, 137 catalogued items". Open it and there are five more folds, one per view, stacked in the article's own column, and each opens inline, in the middle of the long piece it was meant to be an alternative to. A reader with two minutes has to find the fold, open it, find the two-minute version, open that, and read it with the whole article still pressing from below. A reader who only wants the numbers does the same for the catalogue. The views exist; the choice does not, and the views and the article are mixed where they should be side by side.

This article is about the choice. It is a brief for the newsroom, now that the pane on the right of every page is about the piece in front of the reader, and it is also an argument about what not to lose while giving the reader that choice.

In short

• The views worked; the delivery did not. Five readers produced the catalogue, the maps, the arc, the two-minute version and the deck. They arrive as one fold with everything mixed, and a reader short of time is no better off.
• The reader picks the way in, and a default. In the pane about the piece: headlines, the infographic, two minutes, the slides, what is new, the facts, the relationships, the vocabulary, the whole article. One tap. A default per persona. The mode in the address, so it can be sent.
• Deterministic first. Most modes are projections of the article's graph and the views that already exist: computed, explainable, the same for everyone. A model in the reader's own browser is for the rest. Nothing round-trips to the publisher.
• The infographic is often the article. The way the articles are written now, the figures carry most of the story, and a reader who starts there often has what they came for. A figure drawn for one reader, through their own image-model key, is a small step from there.
• What is new is a mode, and it needs the vault. An article that builds on three the reader has read is mostly not new to them. Showing only what it adds takes the Historian's arc plus the reader's own records, which now live in their vault.
• The finished article stays whole, and at the centre. It has a beginning, a sequence, a pacing, a surprise and a voice, and it is the one artefact that has to make sense from end to end. A graph is the right thing for provenance and for zooming in, and it has none of those. The modes are ways in, not replacements.
• Build for the phone first, and for email. The default mode matters most where the screen is smallest.

Focus and time

The problem the five readers were solving is still the right one: the articles here are long because that is where the thinking happens and the evidence lives, and long is hard to consume. The answer was never shorter articles. The answer is that a reader's constraint is time, and their need is focus, and the two change from reader to reader and from day to day. The same person reads the whole thing on a Sunday and wants the headlines on a Tuesday. The same article is a two-minute read for someone who has read the three it builds on and an hour for someone who has not.

So the unit of choice is not the article, it is the way in. And the way in has to be chosen by the reader, at the piece, in a second, which is exactly what the pane about the piece is for.

The modes

Each of these is one tap in the pane. Most of them exist already as a view; what is new is the choice, and the two marked as needing a model.

• Headlines. The title, the In short, and one sentence per section. Computed from the article's structure.
• The infographic. The figures, in order, with their captions. They exist for every article here, and they often carry most of the story: I find, reading my own articles back, that the figure is frequently where I would start. Later, a figure drawn for this reader, from the article's graph, through the reader's own key to an image model, which the bake-off showed costs a few cents.
• Two minutes. The Explainer's view, which exists.
• The slides. The Storyteller's deck, which exists. Later, a short video built from the same slides and the same script, the way code-built video is starting to be made, as a library of pre-made forms rather than a model guessing at pictures.
• What is new. The Historian already writes what an article added and where it sits in the arc. Cross that with what this reader has read, which their vault now records, and the mode shows only the paragraphs and claims that are new to them. For a reader who follows this site, that is the mode I would make the default.
• The facts. The Librarian's catalogue, by kind: the numbers, the claims, the sources, the questions, each anchored to its sentence. Exists.
• The relationships. The article's graph, which exists as JSON beside every article, drawn with its edges to the articles the reader has read highlighted.
• The vocabulary. The Cartographer's ontology, and on demand every term in the article a link, as the inbox brief described.
• At my level, for my persona. A summary pitched to the reader's knowledge, a view through one persona's concerns, a question answered from the piece. This is the mode that needs a model, and it should be the model in the tab: on the reader's machine, with no round trip.
• The whole article. The finished piece, start to end, as written. The centre, below.

Deterministic first, a model second

The split between those two kinds matters more than the list. The views the five readers make are produced once, by agents with role files, and checked; after that, every mode built from them is a projection: computed, explainable, identical for every reader, and free. Headlines, facts, relationships, vocabulary, what is new, the infographic and the slides are all of that kind. They can be built today, they can be tested with a script, and they do not depend on a model being available in the reader's browser.

The mode that needs a model is the one that fits the article to the reader: at their level, through their persona, answering their question. That is what a local model is for, and A model in the tab says where that stands: Chrome's Prompt API ships, the experience is rough, and it will take a while. It does not matter that it takes a while, because the deterministic modes are most of the value, and because when it arrives it runs on the reader's machine. The reader does not depend on the publisher to decide how they read, which is the point of the whole client-side pattern.

What the article keeps that the graph loses

Here is the part I want to be careful about, because it cuts against a habit of mine. Having built the graphs, the catalogues and the views, it is tempting to treat the article as disposable: gather the facts, generate the piece, regenerate it tomorrow. I think that loses something, and I think we should say so.

A finished article has a solidity to it. It is like a book: it has a consistency and a logic, and it is the whole thing. It has a beginning, a sequence of events and a punchline; it introduces an idea and then surprises you with where it goes; its pacing and the position of each fact change what the fact means. It carries the reader on a journey, and it has an author, with a voice. I remember the early output of the writing models, point after point after point, and how hard it was to read, and the reason is that there is a reason we speak in narrative. I read these articles end to end and understand them better for it, and if I printed one and read it with a pen in my hand I would understand it better still. That is how people take in an argument, and nothing about the graph changes it.

The graph loses the narrative. It becomes a set of hyperlinked things, which is exactly right for provenance, for zooming into an area, for learning a concept you did not understand and coming back. It is what an agent reads, and why an agent would pay. But it has no beginning and no voice, and a reader who only ever used it would never have been carried anywhere.

So the finished article is the centre of gravity of everything published, and the rule is this: the article is the one artefact that has to make sense in a coherent way, from end to end, and the graph is the one that has to be checkable. The modes are ways into the article for a reader with twenty seconds, two minutes or an hour. They are not replacements for it, and the design should not let them become one.

One word, while I am here. This site has called the finished article "prose" in several places, and I have come to dislike the word; it sounds like a lesser thing than the graph it stands beside, and it is the opposite. From now on it is the finished article, or the narrative.

The brief

For the session that runs newsroom.sgit.ai. Written by the sgit.ai site session for Dinis Cruz, 12 October 2026, on top of the pane about the piece in The site is an inbox. Dinis decides what ships, and is the first user.

1. The modes in the pane. For an article that has views, a row of modes in the pane about the piece: headlines, infographic, two minutes, slides, what is new, facts, relationships, vocabulary, whole. One tap switches the main column to that mode; the article is never further than one tap away. The fold under the abstract goes.
2. A default, per persona. The reader sets the mode they want to land in, and it is kept with the persona, so the editor persona lands on what is new and the founder persona on headlines.
3. The mode in the address, so a reading list or a link to a colleague can say "the slides of this one".
4. Computed modes first. Headlines, facts, relationships, vocabulary and the infographic from the article's structure, graph and figures; two minutes, the arc and the slides from the views where they exist. Where a view does not exist, the mode says so and offers the nearest one.
5. What is new, from the Historian's arc and the reader's records: the paragraphs and claims this article adds to what the reader has read, with the rest folded. Needs the vault; starts with local storage.
6. The phone. The pane is a sheet from the bottom; the default mode is the first thing shown; the figures are sized for the width; the whole article is one tap.
7. Email. The newsletter cut by the reader's default: headlines, or the infographics, or two minutes per article, with the whole article a link.
8. The model modes, when the model is there. At my level, through my persona, a question about the piece: through the Prompt API when the browser has it, with the page saying plainly when it does not, and never through a server of ours.
9. A figure for this reader, later: the article's graph and a brief to an image model through the reader's own key, as the bake-off did.
10. An experiment per mode. Each mode is small enough to ship in two forms and measure which one a synthetic reader with two minutes gets further with.

What this does not claim

• The views exist for five articles. The modes built from them will exist for those five first; the rest of the site has the graph and the figures, which is enough for headlines, facts, relationships, vocabulary and the infographic.
• The model modes wait for the model. A local model in the browser is rough today, and the design does not depend on it.
• "What is new" needs reading records. It works from local storage on one device and from the vault across devices, and only for a reader who has them.
• Whether a default mode changes how people read is the experiment, and the first reader it will be tested on is me.

If you build reading tools, or you want to read this site in one of these modes before they exist, write to agent@riskmandate.ai.

Where this comes from

A voice memo of mine on 12 October 2026, written by this site's agent in the voice of this site. The argument and the editorial responsibility are mine. The views and their counts are from One article, five readers and the live newsroom on 12 October 2026, where the screenshot of the five readers fold was taken; the cost of a figure from The infographic bake-off; the state of local models from A model in the tab. The figure was drawn for this article. Related: The site is an inbox, The vault is live, The agent is the reader, Story vault underneath, Reader Skills on top and How to run synthetic users.

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles